What problem does it solve? Replicating the exact movement from a real-world reference video onto your own AI-generated characters normally requires manual motion description that never matches the source. This Skill processes a live-action clip locally into a person-depth relief plus OpenPose skeleton control video, binds a 4-15 second segment to a storyboard shot, and sends it as a reference video so the generated clip follows the source motion while the appearance comes from your own character designs. ## Core Features & Use Cases - Local motion extraction: Converts source footage into depth-relief and skeleton control videos entirely on CPU with no API cost, stripping background, faces, and clothing while tracking multiple people separately. - Shot binding and segment framing: Cuts 4-15 second segments with explicit start/end points, aligns shot duration 1:1 with the control segment, and attaches it as a reference video for video generation. - Quality gates and review tooling: Detects ghost tracks and low-detection footage before spending money, enforces motion mode and segment-length rules via lint, and produces three-column source/control/output comparisons after generation. - Use Case: You have a 60-second dance video and want your original character to perform the same choreography. The Skill splits it into five 12-second shots, builds control videos locally, binds each segment, and generates clips where motion follows the source while appearance follows your storyboard images. ## Quick Start Ask the AI to run /kinema-depth with your source video to replicate its motion onto your project characters.
